|Measuring contraceptive use in India: Implications of recent fieldwork design and implementation of the National Family Health Survey|
||Kaushalendra Kumar, Abhishek Singh, and Amy Tsui
||Demographic Research, Volume 47, issue 4; DOI:10.4054/DemRes.2022.47.4
||Background: India’s National Family Health Surveys (NFHS) have provided critical population-level data to inform public policy and research. Although fertility declined, NFHS-4 (2015–2016) reported lower modern contraceptive and female sterilization use compared with NFHS-3 (2005–2006).
Objective: This study assesses selected survey design and interviewer factors’ influences on respondent reporting of modern contraceptive and female sterilization use.
Methods: With data on 582,144 married childbearing-aged females, the analysis pursues multivariable logistic models of both outcomes using survey covariates, assesses interviewer deviance residuals, and estimates multi-level cross-classified random intercept models for state, cluster and interviewer effects.
Results: Adjusted odds ratios (AORs) for reporting modern use in NFHS-4 versus NFHS-3 were 1.21 (1.17–1.26) and 1.66 (1.59–1.74) for sterilization. The AOR for each interview month after survey launch was 1.16 (1.15–1.17) for modern use and 1.18 (1.16–1.19) for sterilization. The AOR for respondents interviewed in the first versus second survey phase was 1.35 (1.30–1.40) for modern methods and 1.12 (1.07–1.17) for female sterilization. Interviewer deviance residuals for both contraceptive outcomes were larger in NFHS-4 than NFHS-3. Eliminating problematic interviews raised modern use 2.0% points and sterilization 1.3% points. Larger state, community cluster and interviewer effects were observed for NFHS-4 versus NFHS-3.
Conclusions: The five-fold expansion of NFHS-4’s sample likely challenged pre-existing survey protocols and may have lowered modern method use by up to 6% points and female sterilization by 2% points.
Contribution: The roles of survey fieldwork and interviewers, as sources of measurement error, are important to consider when interpreting change observed in cross-sectional estimates.